RTFP

In order to support fair and consistent review of potentially violative content, our moderators work alongside our automated moderation systems and take into account additional context and nuance which may not always be picked up by technology. Human moderation also helps improve our automated moderation systems by providing feedback for the underlying machine learning models to strengthen our ongoing detection capabilities. This continuous improvement helps to reduce the volume of potentially distressing videos that moderators view and enables them to focus more on content that requires a greater understanding of context and nuance (such as misinformation, hate speech and harassment). The responsibilities of our content moderators include: Reviewing content flagged by technology: When our automated moderation systems identify potentially problematic content but cannot make an automated decision to remove it, they send the content to our moderation teams for further review. To support this work, we have developed technology that can identify potentially violative items – for example, emblems associated with extremist groups – in video frames, so that content moderators can carefully review the video and the context in which it appears. This technology improves the efficiency of moderators by helping them more adeptly identify violative images or objects, quickly recognise violations, and make decisions accordingly. Reviewing reports from its community: Community reporting helps us maintain a safe environment. We offer our community easy-to-use in-app and online reporting tools so they can flag any content or account they feel is in violation of our Policies or may be illegal. TikTok's trusted partners also play a role in helping detect and remove harmful content. Through our Community Partner Channel Program, participating civil society organizations with a broad range of expertise across online and content safety can report potentially violative or illegal content to us directly for review. Community reports are an important component of our content moderation process. However, the vast majority of removed content is identified proactively before it is reported to us. Reviewing popular content: We manually review video content when it reaches certain levels of popularity in terms of the number of video views, reducing the risk of violative content being shown in the For You Feed or otherwise being widely disseminated. Assessing appeals: If someone disagrees with our decision to restrict or remove their content or an account, they can appeal the decision for reconsideration. These appeals may be sent to moderators to decide if the content should be reinstated on the platform or the account reinstated.

We operate our content moderation processes using a combination of automation and human moderation in accordance with the following four pillars, which provide that we will: Remove content from the platform that violates our Policies; Age-restrict mature content (that does not violate our Community Guidelines but which contains mature themes) so it is only viewed by adults (18 years and older); Maintain For You feed eligibility standards to help ensure any content that may be promoted by the recommendation system is appropriate for a broad audience; and Empower our community with information, tools, and resources.

The tab "9_human_resources" sets out the number of people who are dedicated to content moderation in line with our Policies and applicable local laws, broken down per each of the official European Union languages. Our moderators often have linguistic expertise across multiple languages. Where our moderators have linguistic expertise in more than one European Union language, that expertise is reflected in the detailed language breakdown below. For example, the Czech, Slovakian and Slovenian languages are grouped under one category within our Trust & Safety team and are moderated by the same moderators. The moderators allocated for the Croatian language also cover the Serbian language. These numbers also include moderators covering a number of other languages that are commonly spoken in the region, such as Arabic, Catalan, Hindi, Pashto, Persian, Turkish, Ukrainian, Norwegian, Russian and Icelandic. The numbers in tab 9_human_resources do not reflect the broader teams who also play a key role in keeping our community safe (for example, those involved in the development of our content moderation policies). The total number of human moderators includes non-language specific moderators (meaning the moderators who review profiles or photos).

Qualifications & linguistic expertise: Some of the issues which arise on the platform are highly localised in terms of language and region, which requires deep knowledge and awareness of relevant cultural nuances, terms and context. To address this, and ensure its content moderators are appropriately qualified to make decisions, we have regional policy teams in each region, which includes coverage for all European Union Member States, for example with either designated policy country managers for larger countries or policy managers covering a number of smaller countries. Based primarily in Europe, Regional Policy teams bring regional insights, cultural context, and local expertise to ensure that global moderation policies are applied appropriately across diverse countries and communities in Europe. Acting as expert policy ambassadors, they work to create a safe and positive experience for users by ensuring that our Policies are upheld in ways that reflect local realities. They play a key role in risk mitigation by detecting and addressing regional trends, engaging with external experts such as NGOs, civil society organisations, and government authorities, and collaborating closely with cross-functional teams. Their work includes developing policy interventions and enforcement strategies that strengthen our ability to reduce harm, while maintaining a safe and welcoming environment for our community. The localised policy outputs from the EMEA regional policy team enable our content moderation teams to take a regionally informed approach to content moderation (e.g. rapidly evolving alternative vocabulary or terminology in relation to an unfolding election issue, which may vary/evolve over time and as between countries and languages). We have also established a number of specialised moderation teams to assist our moderators to review content relating to complex issues. For example, assessing harmful misinformation requires additional context and assessment by our misinformation moderators who have enhanced training, expertise and tools to identify such content, including our global repository of previously fact-checked claims from the IFCN-accredited fact-checking partners and direct access to our fact-checking partners where appropriate. We moderate content in more than 70 languages globally and we are transparent in our regular Community Guidelines Enforcement Reports about the primary languages our moderators work in globally. We have language capabilities covering at least one official language for each of the 27 European Union Member States, as well as a number of other languages that are commonly spoken in the region (for example, Arabic and Turkish). This language capability complements our awareness-raising materials, like the Community Guidelines, that are also available in multiple languages. We also have moderation personnel that are not assigned to a particular language, who assist with reviewing content such as photos and profiles.

For this reporting period, TikTok has adopted a revised methodology for assessing the performance of its automated content moderation technologies, in line with the template's requirement in tab "8_automated_means" to report proxy metrics for precision, recall, and accuracy. The three indicators are defined as follows: Precision measures the correctness of the automated system when it takes enforcement action. It is calculated as the proportion of automated enforcement actions that were upheld (i.e., not successfully appealed) out of all automated enforcement actions during the reporting period. Recall measures the completeness of automated detection. It is calculated as upheld automated enforcement actions divided by total upheld enforcement actions across both automated and human detection. Accuracy provides an overall measure of correctness across automated enforcement decisions. It is calculated as upheld automated enforcement actions plus restored human enforcement actions, divided by total enforcement outcomes. This framework relies on two principal ground-truth inputs: User and advertisers appeals, which serve as an indicator of whether an automated enforcement action was correct or erroneous; and Human review outcomes, which capture instances of content that the automated system did not detect but which were subsequently identified and actioned by human moderators. TikTok acknowledges that these inputs function as proxy indicators and that the resulting metrics are subject to the inherent limitations of the underlying data sources. TikTok continues to refine its measurement capabilities and will update its methodology as improved data and analytical frameworks become available.

We place considerable emphasis on proactive detection to remove violative content and reduce exposure to potentially distressing content for our human moderators. Before content is posted to our platform, it's reviewed by automated moderation technologies which identify content or behavior that may violate our policies or For You feed eligibility standards, or that may require age-restriction or other actions. While undergoing this review, the content is visible only to the uploader. If our automated moderation technology identifies content that is a potential violation, it will either take action against the content or flag it for further review by our human moderation teams. In line with our safeguards to help ensure accurate decisions are made, automated removal is applied when violations are the most clear-cut. Some of the methods and technologies that support these efforts include: Vision-based: Computer vision models can identify objects that violate our Policies, such as weapons or hate symbols. Audio-based: Audio clips are reviewed for violations of our Policies, supported by a dedicated audio bank and "classifiers" that help us detect audios that are similar or modified to previous violations. Text-based: Detection models review written content like comments or hashtags, using foundational keyword lists to find variations of violative text. Artificial Intelligence (AI) that can interpret the context surrounding content—helps us identify violations that are context-dependent, such as words that can be used in a hateful way but may not violate our policies by themselves. We also work with various external experts, like our fact-checking partners, to inform our keyword lists. Similarity-based: "Similarity detection systems" enable us to not only catch identical or highly similar versions of violative content, but other types of content that share key contextual similarities and may require additional review. Activity-based: Technologies that look at how accounts are being operated help us disrupt deceptive activities like bot accounts, spam, or attempts to artificially inflate engagement through fake likes or follow attempts. LLMs: We use multimodal LLMs to help moderate content faster and more consistently at scale, from taking automated action on activity like fake engagement, to empowering teams with better moderation tools and risk insights. We work with external groups, for example Tech Against Terrorism in the context of violent extremist content, who help us to more quickly detect and remove violative content that has already been identified off the platform.

If our automated moderation technology identifies content that is a potential violation, it will either take action against the content or flag it for further review by our human moderation teams. In line with our safeguards to help ensure accurate decisions are made, automated removal is applied when violations are the most clear-cut.

Throughout 2025, we have been making ongoing improvements to our safety technologies and enhancements to our content moderation processes. These investments continue to bear out. For example, globally in Q3 2025, we shared our: Highest-ever rate of proactive removal rate (content removed before it's reported to us): 99.3% Highest-ever volume of content removed in under 24 hours: 94.8% Highest-ever rate of violative content removed by automated technologies: 91% If users or advertisers believe we have made a mistake, they can appeal the removal of their content. These advances are helping us remove violative content more quickly, reducing the likelihood of our community seeing it. As more of these tasks are handled by technology, our safety teams can focus more of their time on work that most benefits from human expertise—such as handling appeals, consulting external experts, and responding to fast-moving events. In addition, continued improvements in AI and moderation technologies support the well-being of our safety teams by reducing their exposure to potentially distressing content and equipping them with better tools to carry out this critical work effectively. For example, technologies like AI help make it easier to moderate nuanced areas like misinformation by extracting specific misinformation "claims" from videos for moderators to assess directly or route to our fact-checking partners.

TikTok strives to foster an open and inclusive environment where people can create, find community, and be entertained. To maintain that environment, we take action upon content and accounts that violate our Policies. We are committed to being transparent with our community about the moderation actions we take. The number and type of restrictions we impose as part of our content moderation activities are available in tab "6_own_initiative_TC" Our Policies are the starting point when it comes to how we form and operate our content moderation strategies and practices and they contain provisions which prohibit various forms of illegal and other harmful content. We use a combination of automation and human moderation to identify, review, and action content that violates our Policies.

Support: Human safety professionals continue to play a crucial role in our content moderation approach. We have thousands of safety professionals globally who help build our technologies, develop and enforce our policies, design new safety features, and work with experts and industry peers. We strive to promote a caring working environment for all TikTok employees, and especially for trust and safety professionals. We use an evidence-based approach to develop programs and resources that support their psychological well-being. Our primary focus is on preventative care measures to minimise the risk of psychological injury through well-timed support, training and tools, from recruitment through to onboarding and ongoing employment, that help foster resilience while minimising the risk of psychological injury. These may include tools and features to allow employees to control exposure to graphic content when reviewing or moderating content, including grayscaling, muting and blurring; training for managers to help them identify when a team member may need additional well-being support; and clinical and therapeutic support. We also continue to lean into moderation technology as an effective way to reduce human moderators' exposure to harmful content. We provide our trust and safety employees with membership to the Trust and Safety Professional Association (TSPA). This membership allows them to access resources for career development, participate in workshops and events, and connect with a network of peers across the industry.

Training: To ensure a consistent understanding and application of our Policies, all content moderator personnel receive training across our relevant Policies. All content moderators undergo training on TikTok’s content moderation systems and moderator well-being. Personnel involved in reviewing reported illegal content receive additional focused training on assessing the legality of reported illegal content. Content moderation training materials are kept under review to ensure that they are accurate and current. Such materials include clearly defined learning objectives to ensure our content moderators understand the core policy issues and their underlying policy rationale, key terms and policy exceptions (where applicable). Members of our safety teams attend regular internal sessions dedicated to knowledge sharing and discussion about relevant issues and trends, which include input from external experts. For example, as part of our Election Speaker Series, which helps inform our approach to elections, we invite suitably qualified external local and regional experts to share their insights and market expertise with our internal teams. During the reporting period, we ran 4 Election Speaker Series sessions, 3 in EU Member States, Czechia, Ireland , and Netherlands, and 1 in Moldova. Our teams also participate in various external events to share expertise and support their continued professional learning. These engagements contribute to the team’s awareness of the risks which may arise on the platform, which in turn informs our approach to content moderation.